Learning Bayesian Network Structures to Augment Aircraft Diagnostic Reference Models

Daniel Mack

doi:10.1109/TASE.2016.2542186

Learning Bayesian Network Structures to Augment Aircraft Diagnostic Reference Models

2017-01-01

29

PRE

AI

DOI：10.1109/TASE.2016.2542186

原文链接

原文求助

摘要

En 中文

Fault detection and isolation schemes are designed to detect the onset of adverse events during operations of complex systems, such as aircraft and industrial processes. The state-of-the-art fault diagnosis systems on aircraft combine an expert-created reference model of the associations between faults and symptoms, and a Naive Bayes reasoner. For complex systems with many dependencies between components, the expert-generated reference models are often incomplete, which hinders timely and accurate fault diagnosis. Mining aircraft flight data is a promising approach to finding these missing relations between symptoms and data. However, mining algorithms generate a multitude of relations, and only a small subset of these relations may be useful for improving diagnoser performance. In this paper, we adopt a knowledge engineering approach that combines data mining methods with human expert input to update an existing reference model and improve the overall diagnostic performance. We discuss three case studies to demonstrate the effectiveness of this method. Note to Practitioners-This paper takes a first step toward combining information from adverse event logs matched with real flight data to improve the accuracy and timeliness of diagnoser systems used on commercial aircraft. We have developed a knowledge engineering approach, which uses the results derived from machine learning classifier algorithms to inform experts about changes and additions that could be made to the existing reference model, created by human experts, to improve diagnostic performance. One of the primary constraints we face in this work is not to alter the structure of the diagnostic reference model, which would require changes in the reasoning algorithm for fault diagnosis. With this in mind, we address a number of challenges in developing our methodology. First, we extend the Naive Bayes learning schema by adopting the Tree Augmented Naive Bayesian (TAN) learning algorithm that captures some of the dependencies among the monitors in the aircraft diagnostic system. This provides us with more accurate diagnostic results, and we then apply a transformation schema to generate classifier structures that can be matched against existing reference model structures, thus providing the experts a better understanding of the implications of adding new knowledge and detectors to the reference model. Second, we use real flight data to validate the new reference model structure by determining the improvements in diagnostic accuracy and timeliness of isolation using well-defined metrics. Our overall approach shows promise for targeted fault analysis that may lead to faster detection, and, therefore, avoidance of adverse events such as an engine shutdown during flight. However, the task of studying and refining large, centralized reference models for aircraft systems is complex, especially for quantifying diagnostic accuracy and false alarm rates across multiple fault modes. We will address this larger task along with detection of previously undetected faults (anomaly detection) in future work.

Keyword:

Aviation safety

classification algorithms

diagnosis

knowledge engineering

tree augmented Bayesian networks (TANs)

Learning Bayesian Network Structures to Augment Aircraft Diagnostic Reference Models

摘要

AI总结

期刊

机构

引用论文